AI Forecasting for Enhanced Energy Flexibility in Supermarket Refrigeration Systems

Abstract

Due to the rising misalignment between energy supply and demand, energy flexibility is becoming more relevant. Because supermarket refrigeration systems consume a lot of energy, and their innate ability to act as thermal energy storage, they are a prime candidate to utilize for energy flexibility. This thesis focuses on exploring new parameters which could be used to improve the performance of short term load forecasting models for supermarket refrigeration systems, thereby improving the ability to utilize them for energy flexibility. In addition to this, it compares the performance of multiple different machine learning models. The thesis demonstrates that parameters that have not previously been utilized for short term load forecasting of supermarket refrigeration systems, such as Google’s popular times graph, can successfully be used to improve prediction accuracy.